Transformer secondary winding fault detection method and device based on multispectral signals, terminal equipment and storage medium

Abstract

The invention discloses a transformer secondary winding fault detection method and device based on a multispectral signal, terminal equipment and a storage medium, and relates to the field of power equipment fault detection, and the method comprises the steps: obtaining a first multispectral signal of a to-be-detected transformer, and the first multispectral signal comprises first spectral signals of a plurality of wavebands; calculating an average discharge light pulse amplitude and a discharge light pulse number ratio of each wave band; judging whether the average discharge light pulse amplitudes of all the wavebands are smaller than a preset discharge amplitude or not; if yes, determining that the transformer has no fault; otherwise, determining that the transformer has a fault; and under the condition of determining that the transformer has a fault, calculating the distance between the discharge pulse number ratio and a preset clustering center, and determining the fault condition of the to-be-detected transformer according to the clustering center corresponding to the distance minimum value. By implementing the invention, the problem of misjudgment caused by field sound waves and electromagnetic interference during detection in the prior art can be solved, and accurate identification of the transformer fault is realized.

Description

Technical Field

[0001] This invention relates to the field of power equipment fault detection, and in particular to a method, apparatus, terminal equipment, and storage medium for detecting transformer secondary winding faults based on multispectral signals. Background Technology

[0002] During the operation of power transformers, winding insulation faults frequently occur due to factors such as overvoltage, insulation aging, or operating environment. Common winding insulation faults include partial discharge, flashover, inter-turn short circuit, winding short circuit, or ground fault. Among these, the secondary winding may experience inter-turn discharge, which can progress from a relatively minor fault to the most severe inter-turn floating spark discharge after wire breakage. When the secondary winding experiences a minor inter-turn corona discharge, the transformer can still operate normally. However, when the surface discharge develops to a more severe stage, winding strand breakage and wire disconnection begin to occur. This type of fault will cause a sharp drop in the transformer's output voltage, making it impossible to continue normal operation. As the surface discharge and ablation of the secondary winding continue to develop, the winding insulation will further deteriorate. If appropriate maintenance measures are not taken at this point, the secondary winding will face the risk of complete destruction.

[0003] For secondary winding faults, partial discharge detection methods are currently the primary means of detection and diagnosis. Partial discharge detection methods detect insulation faults by observing acoustic and electrical phenomena, and include ultrasonic methods and pulsed current methods. When an inter-turn short circuit occurs in the winding, changes in the winding's electromagnetic, mechanical, or temperature characteristic parameters can be detected to indicate the fault.

[0004] However, in practical applications, the presence of acoustic interference and complex electromagnetic environments at the site causes a large number of irrelevant interference signals to be mixed into the signals received by traditional electrical methods such as UHF and ultrasonic methods during detection, leading to misjudgment of faults. Summary of the Invention

[0005] This invention provides a method, device, terminal equipment, and storage medium for detecting transformer secondary winding faults based on multispectral signals. It can solve the problem that existing technologies are prone to misjudgment due to on-site acoustic waves and electromagnetic interference when detecting secondary winding faults, and achieve accurate identification of transformer faults.

[0006] One embodiment of the present invention provides a method for detecting transformer secondary winding faults based on multispectral signals, comprising:

[0007] Acquire the first multispectral signal of the transformer under test; the first multispectral signal includes: first spectral signals in several bands;

[0008] Based on the first spectral signal of each band, the average discharge pulse amplitude and the proportion of discharge pulses in each band are calculated.

[0009] Determine whether the average discharge pulse amplitude of all bands is less than the preset discharge amplitude.

[0010] If so, confirm that the transformer under test is fault-free;

[0011] Otherwise, it is determined that the transformer under test is faulty;

[0012] If a fault is found in the transformer under test, the proportion of discharge pulses and the distance between the preset cluster centers are calculated. Based on the cluster center corresponding to the minimum distance, the fault condition of the transformer under test is determined. The fault conditions include: inter-turn corona discharge fault, surface discharge fault, or floating spark discharge fault.

[0013] Furthermore, based on the first spectral signal of each band, the average discharge pulse amplitude and the proportion of discharge pulses in each band are calculated, including:

[0014] Based on the first spectral signal of each band, the amplitude and time of each discharge light pulse in each band are determined by an adaptive thresholding method.

[0015] The average amplitude of the discharge pulses in each band is obtained by summing and averaging the amplitudes of all discharge pulses in each band:

[0016] Obtain the time interval for each applied voltage in the first spectral signal of each band;

[0017] Based on the time of each discharge light pulse in each band and the time interval of each applied voltage, the number of discharge light pulses under each applied voltage in each band is counted.

[0018] The number of discharge light pulses under several applied voltages in each band is summed according to a preset number of times to obtain the dense number of discharge light pulses in each band.

[0019] The discharge pulse density of each band is standardized to obtain the proportion of discharge pulses in each band.

[0020] Furthermore, based on the first spectral signal of each band, the amplitude and time of each discharge light pulse in each band are determined using an adaptive thresholding method, including:

[0021] For each band, the first spectral signal is divided into several sliding window first spectral sub-signals;

[0022] The corresponding background noise is calculated based on the first spectral sub-signal of each sliding window;

[0023] Determine the corresponding adaptive threshold based on the background noise and preset multiplier of each sliding window;

[0024] For each sliding window, if there is an amplitude greater than the adaptive threshold in the first spectral sub-signal of the sliding window, all amplitudes greater than the adaptive threshold in the sliding window are taken as the amplitude of the discharge light pulse, and the time corresponding to the amplitude of the discharge light pulse is taken as the time of the discharge light pulse.

[0025] Furthermore, the pre-defined cluster centers are determined in the following way:

[0026] Acquire several sets of second multispectral signals for a preset fault model; wherein, the fault model includes: inter-turn corona discharge model, surface discharge ablation model and suspended spark discharge model; the second multispectral signal includes: second spectral signals in several bands;

[0027] For each set of second multispectral signals for each fault model, the pulse count ratio of each band is determined based on the second spectral signal of each band, and the pulse count ratio of all bands in the second multispectral signal is used as the spectral data points of the second multispectral signal.

[0028] Clustering is performed on all spectral data points of all fault models to obtain several cluster centers.

[0029] Furthermore, clustering was performed on all spectral data points of all fault models to obtain several cluster centers, including:

[0030] For each set of second multispectral signals for each fault model, the standard deviation of the pulse count ratio for each band is calculated based on the second spectral signal for each band.

[0031] The corresponding dynamic weight is determined based on the standard deviation of the pulse count proportion of each band;

[0032] The number of clusters is determined based on the number of fault models;

[0033] Based on the number of clusters, several spectral data points are randomly selected as the initial cluster centers of the initial clusters.

[0034] Repeat the clustering operation until all spectral data points have been traversed, resulting in several cluster centers;

[0035] Clustering operations include:

[0036] Select an unselected spectral data point from all spectral data points as the current spectral data point;

[0037] Based on the current spectral data points, the current cluster center of each current cluster, and the dynamic weights, calculate the weighted similarity metric between the current spectral data points and each current cluster; where the initial current cluster is the initial cluster, and the initial current cluster center is the initial cluster center.

[0038] The current cluster corresponding to the largest weighted similarity measure value is selected as the target cluster.

[0039] The current spectral data points are assigned to the target cluster to obtain the updated target cluster;

[0040] The current cluster is updated based on the updated target cluster to obtain the updated cluster;

[0041] Based on the updated cluster clusters, recalculate the corresponding updated cluster centers;

[0042] Determine whether all spectral data points have been traversed;

[0043] If the number of spectral data points contained in the updated cluster is greater than the preset number of points, the updated cluster will be used as several cluster centers.

[0044] Otherwise, the updated cluster will be used as the current cluster for the next clustering operation, and the updated cluster center will be used as the current cluster center for the next clustering operation.

[0045] Furthermore, the inter-turn corona discharge model, the surface discharge ablation model, and the suspended spark discharge model are set up in the following ways:

[0046] Obtain defect-free transformer secondary windings;

[0047] The insulation layer between two adjacent turns of enameled wire in the secondary winding of a defect-free transformer is scraped off to obtain an inter-turn corona discharge model.

[0048] The insulation layer of multiple turns of enameled wire on the surface of a defect-free transformer secondary winding is scraped off to obtain a surface discharge ablation model.

[0049] A floating spark discharge model is obtained by cutting one turn of the secondary winding of a defect-free transformer.

[0050] Based on the above method embodiments, the present invention provides corresponding device embodiments, including: a multispectral data acquisition module, a discharge light pulse parameter calculation module, a light pulse amplitude judgment module, a fault-free judgment module, a fault judgment module, and a fault type identification module;

[0051] The multispectral data acquisition module is used to acquire the first multispectral signal of the transformer under test; the first multispectral signal includes: first spectral signals in several bands;

[0052] The discharge pulse parameter calculation module is used to calculate the average discharge pulse amplitude and the proportion of discharge pulses in each band based on the first spectral signal of each band.

[0053] The optical pulse amplitude judgment module is used to determine whether the average discharge optical pulse amplitude of all bands is less than the preset discharge amplitude.

[0054] The fault determination module is used to determine that the transformer under test is fault-free if the condition is met.

[0055] The fault determination module is used to determine, otherwise, that the transformer under test has a fault.

[0056] The fault type identification module is used to calculate the proportion of discharge pulses and the distance between preset cluster centers when it is determined that there is a fault in the transformer under test. Based on the cluster center corresponding to the minimum distance, the fault condition of the transformer under test is determined. The fault conditions include: inter-turn corona discharge fault, surface discharge fault, or floating spark discharge fault.

[0057] Furthermore, the pre-defined cluster centers are determined in the following way:

[0058] Acquire several sets of second multispectral signals for a preset fault model; wherein, the fault model includes: inter-turn corona discharge model, surface discharge ablation model and suspended spark discharge model; the second multispectral signal includes: second spectral signals in several bands;

[0059] For each set of second multispectral signals for each fault model, the pulse count ratio of each band is determined based on the second spectral signal of each band, and the pulse count ratio of all bands in the second multispectral signal is used as the spectral data points of the second multispectral signal.

[0060] Clustering is performed on all spectral data points of all fault models to obtain several cluster centers.

[0061] Based on the above method embodiments, the present invention provides a corresponding terminal device embodiment, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps of the transformer secondary winding fault detection method based on multispectral signals as described in the present invention.

[0062] Based on the above method embodiments, the present invention provides a corresponding computer-readable storage medium embodiment, including: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform the steps of the transformer secondary winding fault detection method based on multispectral signals as described in the present invention.

[0063] Compared with the prior art, the beneficial effects of this embodiment are as follows:

[0064] This invention acquires a first multispectral signal of the transformer under test, which includes first spectral signals in several bands. Since different bands of spectral signals have different response characteristics to different physical phenomena and fault characteristics inside the transformer, and spectral signals themselves are not easily affected by acoustic waves and electromagnetic interference, this invention utilizes multi-band spectral information to improve the anti-interference capability of fault detection. Based on the first spectral signal of each band, the average discharge light pulse amplitude and the proportion of discharge light pulses in each band are calculated. These two parameters reflect the internal discharge situation of the transformer from different perspectives. The average discharge light pulse amplitude reflects the intensity characteristics of the internal discharge of the transformer, while the proportion of discharge light pulses reflects the frequency of discharge occurrence. Determine if the average discharge pulse amplitude of all bands is less than the preset discharge amplitude. If so, it indicates that the internal discharge intensity of the transformer is weak, and the transformer under test is determined to be fault-free. Otherwise, the transformer under test is determined to be faulty. If the transformer under test is determined to be faulty, calculate the distance between the proportion of discharge pulses and the preset cluster center. The smaller the distance, the more similar the fault characteristics of the transformer under test are to the fault type represented by the cluster center. Based on the cluster center corresponding to the minimum distance, determine the fault condition of the transformer under test. Fault conditions include: inter-turn corona discharge fault, surface discharge fault, or floating spark discharge fault.

[0065] In summary, this invention utilizes the inherent resistance of multi-band spectral data to acoustic and electromagnetic interference, and combines it with cluster analysis for accurate fault detection. This solves the problem of misjudgment caused by on-site acoustic and electromagnetic interference in secondary winding fault detection in existing technologies, and achieves accurate identification of transformer faults. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating a transformer secondary winding fault detection method based on multispectral signals according to an embodiment of the present invention.

[0067] Figure 2 This is a curve showing the center wavelength and transmittance characteristics of each filter provided in an embodiment of the present invention;

[0068] Figure 3 This is a curve showing the variation of the proportion of discharge pulses in each band of inter-turn corona discharge according to an embodiment of the present invention;

[0069] Figure 4 This is a curve showing the variation of the proportion of discharge pulses in each band of surface discharge according to an embodiment of the present invention;

[0070] Figure 5 This is a curve showing the variation of the proportion of discharge pulses in each band of suspended spark discharge according to an embodiment of the present invention;

[0071] Figure 6 This is a schematic diagram of the structure of a transformer secondary winding fault detection device based on multispectral signals provided in an embodiment of the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated.

[0074] like Figure 1 As shown, in order to solve the problem that existing technologies are prone to misjudgment due to on-site acoustic waves and electromagnetic interference when detecting secondary winding faults, an embodiment of the present invention provides a transformer secondary winding fault detection method based on multispectral signals. The method includes at least the following steps:

[0075] Step S1: Acquire the first multispectral signal of the transformer under test; the first multispectral signal includes: first spectral signals of several bands;

[0076] For step S1, a first multispectral signal of the transformer under test is captured by a multispectral sensor. The first multispectral signal includes first spectral signals in several bands. In order to comprehensively cover possible abnormal signals of the transformer, such as ultraviolet light generated by partial discharge, visible light reflected by surface defects, and near-infrared light associated with internal insulation aging, in this embodiment, the first multispectral signal includes first spectral signals in three bands, corresponding to the ultraviolet band, visible light band, and near-infrared band, respectively.

[0077] The multispectral sensor includes a multispectral photosensitive array, a filter, and a bracket for fixing the filter. The multispectral photosensitive array uses a single-photon avalanche diode (SiPM) array with a photosensitive area of ​​6mm×6mm. Multiple units are arranged in a square to form a 2×2 photosensitive array. This array structure can expand the detection range and ensure that even the weak light signal in the early stage of a transformer fault can be accurately captured thanks to the high sensitivity of SiPM.

[0078] In this invention, the operating spectral band of the filter is 200–800 nm. Preferably, the filter uses a bandpass filter with center wavelengths of 310 nm, 500 nm, and 700 nm, enabling the SiPM array to capture and detect components in the ultraviolet, visible, and near-infrared bands. Figure 2 The figure shows the center wavelength and transmittance characteristics of each filter.

[0079] Step S2: Based on the first spectral signal of each band, calculate the average discharge pulse amplitude and the proportion of discharge pulses in each band;

[0080] In a preferred embodiment, based on the first spectral signal of each band, the average discharge pulse amplitude and the proportion of discharge pulses in each band are calculated, including:

[0081] Based on the first spectral signal of each band, the amplitude and time of each discharge light pulse in each band are determined by an adaptive thresholding method.

[0082] The average amplitude of the discharge pulses in each band is obtained by summing and averaging the amplitudes of all discharge pulses in each band:

[0083] Obtain the time interval for each applied voltage in the first spectral signal of each band;

[0084] Based on the time of each discharge light pulse in each band and the time interval of each applied voltage, the number of discharge light pulses under each applied voltage in each band is counted.

[0085] The number of discharge light pulses under several applied voltages in each band is summed according to a preset number of times to obtain the dense number of discharge light pulses in each band.

[0086] The discharge pulse density of each band is standardized to obtain the proportion of discharge pulses in each band.

[0087] For step S2, firstly, the threshold value is automatically adjusted according to the characteristics of the first spectral signal in each band using an adaptive thresholding method, thereby accurately identifying each discharge light pulse in the first spectral signal. For each identified discharge light pulse, its two key parameters, amplitude and time, are further determined. The amplitude reflects the intensity of the discharge light pulse, while the time records the specific moment when the discharge light pulse appears.

[0088] Preferably, based on the first spectral signal of each band, the amplitude and time of each discharge light pulse in each band are determined by an adaptive thresholding method, including:

[0089] For each band, the first spectral signal is divided into several sliding window first spectral sub-signals;

[0090] The corresponding background noise is calculated based on the first spectral sub-signal of each sliding window;

[0091] Determine the corresponding adaptive threshold based on the background noise and preset multiplier of each sliding window;

[0092] For each sliding window, if there is an amplitude greater than the adaptive threshold in the first spectral sub-signal of the sliding window, all amplitudes greater than the adaptive threshold in the sliding window are taken as the amplitude of the discharge light pulse, and the time corresponding to the amplitude of the discharge light pulse is taken as the time of the discharge light pulse.

[0093] Specifically, for the processing of the first spectral signal in each band, the first spectral signal is first divided into several consecutive sliding windows, each window corresponding to an independent first spectral sub-signal. Within each sliding window, the background noise of the first spectral sub-signal within each sliding window is calculated in real time, and an adaptive threshold is set based on the background noise and a preset multiple t. The adaptive threshold is set to t times the background noise, where the value of t needs to be selected according to the actual situation.

[0094] Within the current sliding window, by comparing the amplitude of the first spectral sub-signal with the aforementioned adaptive threshold, it is determined whether a signal with an amplitude greater than the adaptive threshold exists. If a signal with an amplitude exceeding the adaptive threshold exists within the window, it indicates that a discharge optical pulse related to a transformer fault may have been captured within that window. In this case, all amplitudes greater than the adaptive threshold within the sliding window are taken as the amplitudes of the discharge optical pulses. Simultaneously, the time corresponding to the amplitude of the discharge optical pulse is taken as the time of the discharge optical pulse. After completing the analysis of the current window, the process continues to traverse the next sliding window. If no signal with an amplitude exceeding the adaptive threshold exists within the window, it indicates that no valid discharge optical pulse has been detected in that window, and the process continues to traverse the next sliding window. After all sliding windows have been traversed, the amplitude and time of each discharge optical pulse within the current band are obtained.

[0095] After obtaining the amplitude and time of each discharge light pulse in each band, in order to intuitively reflect the intensity of the discharge phenomenon in different bands, a statistical analysis is performed on the amplitude of all effective discharge light pulses in each band. Specifically, the average amplitude of the discharge light pulses in each band is obtained by summing and averaging the amplitudes of all discharge light pulses in each band using the following formula:

[0096]

[0097] Among them, M kmean M represents the average discharge light pulse amplitude in the k-th band, where k represents the band index. ki This represents the amplitude of the i-th discharge pulse in the k-th band, where i represents the index of the discharge pulse, i = 1, 2, ..., n. k n k This represents the total number of discharge light pulses in the k-th band.

[0098] After obtaining the average discharge light pulse amplitude for each band, in order to further analyze the distribution characteristics of the discharge phenomenon, the proportion of the number of discharge light pulses in each band is calculated, thereby reflecting the occurrence pattern and proportion of the discharge phenomenon in different bands within the voltage application period, and quantifying the frequency of discharge occurrence.

[0099] Specifically, firstly, the time interval of each voltage application in the first spectral signal of each band is obtained to clarify the time range of voltage application. Based on the time of each discharge light pulse in each band and the time interval of each voltage application, the number of discharge light pulses under each voltage application in each band is counted to obtain the discharge frequency of a single voltage application. Then, the number of discharge light pulses under several voltage applications in each band is summed according to a preset number of times to obtain the discharge light pulse density of each band, realizing the statistics of the total discharge under multiple voltage applications. The specific formula is as follows:

[0100]

[0101] Where, N k This represents the density of discharge light pulses in the k-th band, where H represents the preset number of times. In this embodiment, H = 100, and N... k,j This represents the number of discharge light pulses in the k-th band under the j-th voltage, where j represents the index of the number of times the voltage was applied.

[0102] Finally, the discharge pulse density number of each band is standardized using the following formula to obtain the proportion of discharge pulses in each band:

[0103]

[0104] Where, N k % represents the percentage of discharge light pulses in the k-th band, and L represents the number of band signals in the multispectral signal. In this embodiment, L = 3.

[0105] It should be noted that, after standardization, the sum of the percentages of discharge light pulses across all bands is always 1, i.e.

[0106] Step S3: Determine whether the average discharge pulse amplitude of all bands is less than the preset discharge amplitude;

[0107] For step S3, the average amplitude obtained for each band is compared with the preset discharge amplitude to determine whether the average discharge light pulse amplitude of all bands is less than the preset discharge amplitude, thereby evaluating the overall discharge state of the transformer.

[0108] It should be noted that the preset discharge amplitude is a key threshold for measuring the discharge intensity, which is set in advance based on the specific conditions such as the transformer's design parameters. In this embodiment, the preset discharge amplitude is set to 1.2 to 1.5 times the background noise. Here, 1.2 times is used, but different values ​​can be selected according to the actual situation.

[0109] Step S4: If yes, confirm that the transformer under test is fault-free;

[0110] For step S4, after the judgment in step S3, if the average discharge light pulse amplitude of all bands is less than the preset discharge amplitude, it means that the transformer under test has not produced obvious discharge characteristics in each band, that is, the current discharge behavior is within the controllable range, and it is determined that the transformer under test is fault-free.

[0111] Preferably, once it is determined that the transformer under test is fault-free, regular maintenance is performed on the transformer.

[0112] Step S5: Otherwise, determine that the transformer under test is faulty;

[0113] For step S5, after the judgment in step S3, if the average discharge light pulse amplitude of at least one band is greater than or equal to the preset discharge amplitude, it indicates that an abnormal discharge has occurred in the band range, and it is determined that the transformer under test has a fault.

[0114] Step S6: If it is determined that there is a fault in the transformer under test, calculate the proportion of discharge pulses and the distance between the preset cluster centers, and determine the fault condition of the transformer under test according to the cluster center corresponding to the minimum distance; the fault conditions include: inter-turn corona discharge fault, surface discharge fault, or floating spark discharge fault.

[0115] For step S6, after step S5 confirms that the transformer under test has a fault, different types of discharge faults will exhibit different distribution characteristics of the proportion of discharge pulses across multiple frequency bands. This characteristic can be used as a basis for distinguishing fault types. Therefore, to further determine the fault type, the distance between the proportion of discharge pulses and preset cluster centers is calculated to quantify the similarity between the current fault characteristics and various preset fault types. A smaller distance means that the current fault characteristics are closer to the fault type represented by the cluster center. Finally, the calculated proportion of discharge pulses of the transformer under test is compared and analyzed with these preset cluster centers. Based on the principle of minimum distance, it can be determined whether the transformer has an inter-turn corona discharge fault, a surface discharge fault, or a floating spark discharge fault, thus providing a clear direction for transformer fault diagnosis and subsequent processing.

[0116] Preferably, when an inter-turn corona discharge fault is confirmed in the transformer under test, if the inter-turn corona discharge persists, it will gradually erode the insulation material, reduce insulation performance, and may lead to serious faults such as inter-turn short circuits. Repair should be arranged promptly. When a surface discharge fault is confirmed in the transformer under test, surface discharge will continuously damage the insulation surface, further deteriorating insulation performance. If not addressed promptly, it may lead to insulation breakdown, threatening the insulation safety of the secondary winding. A warning should be issued to prevent further deterioration of the secondary winding insulation. When a floating spark discharge fault is confirmed in the transformer under test, floating spark discharge will generate high temperatures and strong electromagnetic interference, causing serious damage to the internal insulation material and metal structure of the transformer. It is highly likely to cause more serious faults in a short period of time. The transformer should be shut down for maintenance promptly to prevent serious faults.

[0117] It should be noted that the preset cluster centers are determined in advance based on a large amount of experimental data and fault characteristic studies. Specifically, they cover the cluster centers of the discharge pulse proportions of three types of discharge—inter-turn discharge, surface discharge, and floating discharge—in three wavelength bands, namely c1, c2, and c3. In this embodiment, c1 = (0.15, 0.6, 0.25), representing the discharge pulse proportion of inter-turn discharge in the ultraviolet, visible, and near-infrared bands; c2 = (0.25, 0.55, 0.20), representing the discharge pulse proportion of surface discharge in the ultraviolet, visible, and near-infrared bands; and c3 = (0.2, 0.5, 0.3), representing the discharge pulse proportion of floating discharge in the ultraviolet, visible, and near-infrared bands.

[0118] Next, the process of constructing the preset cluster centers will be explained in detail:

[0119] In a preferred embodiment, the preset cluster centers are determined in the following manner:

[0120] Acquire several sets of second multispectral signals for a preset fault model; wherein, the fault model includes: inter-turn corona discharge model, surface discharge ablation model and suspended spark discharge model; the second multispectral signal includes: second spectral signals in several bands;

[0121] For each set of second multispectral signals for each fault model, the pulse count ratio of each band is determined based on the second spectral signal of each band, and the pulse count ratio of all bands in the second multispectral signal is used as the spectral data points of the second multispectral signal.

[0122] Clustering is performed on all spectral data points of all fault models to obtain several cluster centers.

[0123] In a preferred embodiment, the inter-turn corona discharge model, the surface discharge ablation model, and the suspended spark discharge model are configured in the following manner:

[0124] Obtain defect-free transformer secondary windings;

[0125] The insulation layer between two adjacent turns of enameled wire in the secondary winding of a defect-free transformer is scraped off to obtain an inter-turn corona discharge model.

[0126] The insulation layer of multiple turns of enameled wire on the surface of a defect-free transformer secondary winding is scraped off to obtain a surface discharge ablation model.

[0127] A floating spark discharge model is obtained by cutting one turn of the secondary winding of a defect-free transformer.

[0128] In one embodiment of the present invention, firstly, three types of fault models are constructed by targeted treatment of defect-free transformer secondary windings. Specifically, the insulation layer between two adjacent turns of enameled wire is scraped off to obtain an inter-turn corona discharge model; the insulation layer of multiple consecutive turns of enameled wire on the winding surface is scraped off to obtain a surface discharge ablation model; and one turn of the coil is cut off to obtain a suspended spark discharge model.

[0129] Subsequently, these secondary winding models with discharge defects were installed in the simulation cavity and electrically connected. The fault light signals generated by these three types of fault models were detected in the ultraviolet, visible, and near-infrared spectral bands using multispectral sensors, thereby acquiring several sets of second multispectral signals to provide basic samples for the subsequent determination of cluster centers.

[0130] In this embodiment, to ensure the accuracy and robustness of the cluster centers, 20,000 sets of second multispectral signals are collected for each type of fault model. Since there are 3 types of fault models, a total of 3×20,000 sets of second multispectral signals are acquired. Each set of second multispectral signals contains spectral signals in three bands: ultraviolet, visible, and near-infrared, so as to comprehensively cover the characteristic performance of different fault types in the multispectral dimension.

[0131] Next, for each set of second multispectral signals in each fault model, the same processing logic as for the first multispectral signal is used. First, for the second spectral signals in the ultraviolet, visible and near-infrared bands, the amplitude and time of each discharge light pulse in each band are determined by the adaptive threshold method. Then, the number of pulses under a single voltage action is counted by combining the time interval of each applied voltage. Then, the discharge light pulse density of each band is accumulated according to the preset number of times. Finally, the proportion of discharge light pulses in each band is obtained by standardization.

[0132] After calculating the proportions of the three bands, this set of proportion data, such as the proportion of ultraviolet band x, the proportion of visible light band y, and the proportion of near-infrared band z, is integrated into a three-dimensional feature vector, which is used as the spectral data point corresponding to the second multispectral signal.

[0133] In order to extract typical spectral feature patterns of various faults from massive feature data, it is necessary to perform cluster analysis on all spectral data points of all fault models, and then obtain several cluster centers that correspond one-to-one with different fault types.

[0134] In a preferred embodiment, all spectral data points of all fault models are clustered to obtain several cluster centers, including:

[0135] For each set of second multispectral signals for each fault model, the standard deviation of the pulse count ratio for each band is calculated based on the second spectral signal for each band.

[0136] The corresponding dynamic weight is determined based on the standard deviation of the pulse count proportion of each band;

[0137] The number of clusters is determined based on the number of fault models;

[0138] Based on the number of clusters, several spectral data points are randomly selected as the initial cluster centers of the initial clusters.

[0139] Repeat the clustering operation until all spectral data points have been traversed, resulting in several cluster centers;

[0140] Clustering operations include:

[0141] Select an unselected spectral data point from all spectral data points as the current spectral data point;

[0142] Based on the current spectral data points, the current cluster center of each current cluster, and the dynamic weights, calculate the weighted similarity metric between the current spectral data points and each current cluster; where the initial current cluster is the initial cluster, and the initial current cluster center is the initial cluster center.

[0143] The current cluster corresponding to the largest weighted similarity measure value is selected as the target cluster.

[0144] The current spectral data points are assigned to the target cluster to obtain the updated target cluster;

[0145] The current cluster is updated based on the updated target cluster to obtain the updated cluster;

[0146] Based on the updated cluster clusters, recalculate the corresponding updated cluster centers;

[0147] Determine whether all spectral data points have been traversed;

[0148] If the number of spectral data points contained in the updated cluster is greater than the preset number of points, the updated cluster will be used as several cluster centers.

[0149] Otherwise, the updated cluster will be used as the current cluster for the next clustering operation, and the updated cluster center will be used as the current cluster center for the next clustering operation.

[0150] In one embodiment of the present invention, firstly, for each set of second multispectral signals for each fault model, the standard deviation of the pulse count ratio for each band is calculated using the following formula based on the second spectral signal for each band:

[0151]

[0152] Where, σ d This represents the standard deviation of the pulse count percentage of the d-th band in the second multispectral signal, where d represents the band index in the second multispectral signal, and n... d N represents the number of discharge light pulses in the d-th band of the second multispectral signal. di This represents the percentage of the number of discharge light pulses in the ith band of the second multispectral signal, where i ∈ {1, 2, ..., n}. d}, N dmean This indicates that the nth band in the second multispectral signal represents the dth band. d The average percentage of each discharge light pulse.

[0153] It should be noted that the standard deviation of the pulse number percentage reflects the degree of dispersion of the data in that band. Different degrees of dispersion mean that there are differences in the stability and concentration of the data characteristics.

[0154] To make more rational use of the data from each band, the reciprocal of the standard deviation of the pulse number proportion for each band is taken to obtain the corresponding dynamic weight. The specific formula is as follows:

[0155]

[0156] Among them, w d This represents the dynamic weight of the d-th band in the second multispectral signal.

[0157] Next, the number of clusters is determined based on the number of fault models. In this embodiment, the fault models include inter-turn corona discharge model, surface discharge ablation model, and suspended spark discharge model, and the corresponding number of clusters K is set to 3. It should be noted that matching the number of clusters with the number of fault models ensures the relevance and accuracy of subsequent cluster analysis.

[0158] Based on the number of clusters, several spectral data points are randomly selected as the initial cluster centers of the initial clusters. In this embodiment, K spectral data points are randomly selected without repetition from all spectral data points as the initial cluster centers C = {x1, x2, ..., x...}. K}

[0159] Then, the clustering operation is repeated until all spectral data points have been traversed, resulting in several cluster centers. Specifically, the clustering operation first selects an unselected spectral data point from all spectral data points as the current spectral data point.

[0160] Based on the current spectral data points, the current cluster centers of each current cluster, and the dynamic weights, the weighted similarity metric between the current spectral data points and each current cluster is calculated using the following formula:

[0161]

[0162] Where, x u c represents the u-th spectral data point, i.e., a different current spectral data point. v Let s(x) represent the current cluster center of the v-th current cluster. u ,c v () represents the weighted similarity metric between the u-th spectral data point and the v-th current cluster, d represents the index of the band in the second multispectral signal, which is also the index of the dimension of the spectral data point, and D represents the total number of bands in the second multispectral signal, which is also the total number of dimensions of the spectral data point. In this embodiment, the bands in the second multispectral signal include ultraviolet, visible, and near-infrared, i.e., D = 3, x u,d c represents the d-th dimension value of the u-th spectral data point in the second multispectral signal. v,d This represents the d-th dimension value of the current cluster center of the v-th current cluster in the second multispectral signal.

[0163] It should be noted that the current cluster at the beginning is the initial cluster cluster, and the current cluster center at the beginning is the initial cluster center.

[0164] The weighted similarity measure reflects the degree of similarity between the current spectral data point and each cluster. The larger the value, the higher the similarity. By selecting the current cluster corresponding to the largest weighted similarity measure as the target cluster, it can be ensured that the current spectral data point is assigned to the cluster with the closest features. Adding the current spectral data point to the target cluster changes the set of data points contained in the cluster, thus obtaining the updated target cluster.

[0165] At this point, the updated target cluster replaces the corresponding original cluster state in the current cluster, resulting in updated clusters. Based on the updated clusters, the corresponding updated cluster centers are recalculated to represent the feature positions of each cluster. The specific formula is as follows:

[0166]

[0167] Among them, c v ′ C represents the cluster center corresponding to the target cluster. v Let s(x) represent the v-th updated cluster. u C v ) represents the weighted similarity measure between the u-th spectral data point and the v-th updated cluster, and w represents the feature weight vector. In this embodiment, w = (w1, w2, w3).

[0168] Determine whether the corresponding clustering operation has been performed for each spectral data point. If there are still spectral data points that have not been traversed, the clustering operation needs to be performed again until all spectral data points are included in the corresponding clusters.

[0169] If it is confirmed that all spectral data points have been traversed, further determine whether the number of spectral data points contained in the updated cluster is greater than the preset number of points. If the number of spectral data points contained in the updated cluster is greater than the preset number of points, it means that the updated cluster has sufficient representativeness and scale, and the updated cluster is used as the corresponding cluster center. If the number of spectral data points contained in the updated cluster is not greater than the preset number of points, it means that the scale of the cluster is relatively small, which may lead to certain limitations and randomness in the characteristics reflected by the cluster, and it cannot fully represent the general characteristics of a class of data. It is marked as a noise cluster, and in this case, it is necessary to reselect the initial cluster center and re-perform the clustering operation.

[0170] Repeat the above clustering operations to continuously optimize the clustering results and gradually improve the accuracy and reliability of clustering, so that the final cluster centers can accurately reflect the inherent structure and characteristics of the data.

[0171] In this invention, such as Figure 3 The figure shows the variation curves of the proportion of discharge pulses in each band of inter-turn corona discharge. The trend of the proportion of discharge pulses is more obvious. With the increase of discharge times, the proportion of discharge pulses in the 310nm and 700nm bands increases, while the proportion of discharge pulses in the 500nm band is exactly the opposite. When the number of discharges exceeds 15,000, the proportion of pulses in each band no longer changes significantly.

[0172] like Figure 4The figure shows the variation curve of the proportion of discharge pulses in each band of surface discharge. As the number of triggers increases, the proportion of discharge pulses in the 310nm and 700nm bands is relatively stable before 15,000 triggers. After about 15,000 triggers, the proportion of pulses in the two bands increases, while the proportion of discharge pulses in the 500nm band begins to decrease. The three bands show the characteristic of "stabilizing first and then changing" as the number of triggers increases, which is exactly the opposite of the characteristic of "changing first and then stabilizing" of inter-turn corona discharge.

[0173] like Figure 5 The figure shows the variation curves of the proportion of discharge pulses in each band of suspended spark discharge. The stability of the proportion of discharge pulses in each band is significantly higher than that of inter-turn corona discharge and surface discharge. The proportion of discharge pulses in the 700nm band is higher than that of the two discharge types mentioned above. The proportion of discharge pulses in the 310nm band is slightly higher than that of inter-turn corona discharge faults, but slightly lower than that of surface discharge faults. The proportion of discharge pulses in the 500nm band is roughly the same as that of surface discharge faults, and slightly lower than that of inter-turn corona discharge faults.

[0174] like Figure 6 As shown, based on the above method embodiments, corresponding apparatus embodiments are provided;

[0175] An embodiment of the present invention provides a transformer secondary winding fault detection device based on multispectral signals, comprising: a multispectral data acquisition module, a discharge optical pulse parameter calculation module, an optical pulse amplitude judgment module, a fault-free judgment module, a fault judgment module, and a fault type identification module;

[0176] The multispectral data acquisition module is used to acquire the first multispectral signal of the transformer under test; the first multispectral signal includes: first spectral signals in several bands;

[0177] The discharge pulse parameter calculation module is used to calculate the average discharge pulse amplitude and the proportion of discharge pulses in each band based on the first spectral signal of each band.

[0178] The optical pulse amplitude judgment module is used to determine whether the average discharge optical pulse amplitude of all bands is less than the preset discharge amplitude.

[0179] The fault determination module is used to determine that the transformer under test is fault-free if the condition is met.

[0180] The fault determination module is used to determine, otherwise, that the transformer under test has a fault.

[0181] The fault type identification module is used to calculate the proportion of discharge pulses and the distance between preset cluster centers when it is determined that there is a fault in the transformer under test. Based on the cluster center corresponding to the minimum distance, the fault condition of the transformer under test is determined. The fault conditions include: inter-turn corona discharge fault, surface discharge fault, or floating spark discharge fault.

[0182] In a preferred embodiment, the preset cluster centers are determined in the following manner:

[0183] Acquire several sets of second multispectral signals for a preset fault model; wherein, the fault model includes: inter-turn corona discharge model, surface discharge ablation model and suspended spark discharge model; the second multispectral signal includes: second spectral signals in several bands;

[0184] For each set of second multispectral signals for each fault model, the pulse count ratio of each band is determined based on the second spectral signal of each band, and the pulse count ratio of all bands in the second multispectral signal is used as the spectral data points of the second multispectral signal.

[0185] Clustering is performed on all spectral data points of all fault models to obtain several cluster centers.

[0186] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the transformer secondary winding fault detection method based on multispectral signals provided by any of the above-described method embodiments of the present invention.

[0187] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0188] Based on the above embodiments of the transformer secondary winding fault detection method based on multispectral signals, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the transformer secondary winding fault detection method based on multispectral signals according to any embodiment of the present invention.

[0189] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0190] The terminal device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and memory.

[0191] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting various parts of the terminal device via various interfaces and lines.

[0192] Based on the above-described method embodiments, another embodiment is provided: another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the transformer secondary winding fault detection method based on multispectral signals as described in any of the above-described method embodiments of the present invention.

[0193] The module / unit integrated into the transformer secondary winding fault detection device / terminal equipment based on multispectral signals, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0194] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for detecting faults in the secondary winding of a transformer based on multispectral signals, characterized in that, include: Acquire the first multispectral signal of the transformer under test; The first multispectral signal includes: a first spectral signal in several bands; Based on the first spectral signal of each band, the average discharge pulse amplitude and the proportion of discharge pulses in each band are calculated. Determine whether the average discharge pulse amplitude of all bands is less than the preset discharge amplitude. If so, confirm that the transformer under test is fault-free; Otherwise, it is determined that the transformer under test is faulty; If a fault is found in the transformer under test, the proportion of discharge pulses and the distance between the preset cluster centers are calculated, and the fault condition of the transformer under test is determined according to the cluster center corresponding to the minimum distance. The fault conditions include: inter-turn corona discharge fault, surface discharge fault, or floating spark discharge fault.

2. The method for detecting transformer secondary winding faults based on multispectral signals according to claim 1, characterized in that, Based on the first spectral signal of each band, the average discharge pulse amplitude and the proportion of discharge pulses in each band are calculated, including: Based on the first spectral signal of each band, the amplitude and time of each discharge light pulse in each band are determined by an adaptive thresholding method. The average amplitude of the discharge pulses in each band is obtained by summing and averaging the amplitudes of all discharge pulses in each band: Obtain the time interval for each applied voltage in the first spectral signal of each band; Based on the time of each discharge light pulse in each band and the time interval of each applied voltage, the number of discharge light pulses under each applied voltage in each band is counted. The number of discharge light pulses under several applied voltages in each band is summed according to a preset number of times to obtain the dense number of discharge light pulses in each band. The discharge pulse density of each band is standardized to obtain the proportion of discharge pulses in each band.

3. The transformer secondary winding fault detection method based on multispectral signals according to claim 2, characterized in that, Based on the first spectral signal of each band, the amplitude and time of each discharge light pulse in each band are determined by an adaptive thresholding method, including: For each band, the first spectral signal is divided into several sliding window first spectral sub-signals; The corresponding background noise is calculated based on the first spectral sub-signal of each sliding window; Determine the corresponding adaptive threshold based on the background noise and preset multiplier of each sliding window; For each sliding window, if there is an amplitude greater than the adaptive threshold in the first spectral sub-signal of the sliding window, all amplitudes greater than the adaptive threshold in the sliding window are taken as the amplitude of the discharge light pulse, and the time corresponding to the amplitude of the discharge light pulse is taken as the time of the discharge light pulse.

4. The transformer secondary winding fault detection method based on multispectral signals according to claim 1, characterized in that, The preset cluster centers are determined in the following way: Acquire several sets of second multispectral signals of a preset fault model; wherein, the fault model includes: an inter-turn corona discharge model, a surface discharge ablation model, and a suspended spark discharge model; the second multispectral signal includes: second spectral signals of several bands; For each set of second multispectral signals for each fault model, the pulse count ratio of each band is determined based on the second spectral signal of each band, and the pulse count ratio of all bands in the second multispectral signal is used as the spectral data points of the second multispectral signal. Clustering is performed on all spectral data points of all fault models to obtain several cluster centers.

5. The transformer secondary winding fault detection method based on multispectral signals according to claim 4, characterized in that, Clustering was performed on all spectral data points of all fault models to obtain several cluster centers, including: For each set of second multispectral signals for each fault model, the standard deviation of the pulse count ratio for each band is calculated based on the second spectral signal for each band. The corresponding dynamic weight is determined based on the standard deviation of the pulse count proportion of each band; The number of clusters is determined based on the number of fault models; Based on the number of clusters, several spectral data points are randomly selected as the initial cluster centers of the initial clusters. Repeat the clustering operation until all spectral data points have been traversed, resulting in several cluster centers; The clustering operation includes: Select an unselected spectral data point from all spectral data points as the current spectral data point; Based on the current spectral data points, the current cluster center of each current cluster, and the dynamic weights, calculate the weighted similarity metric between the current spectral data points and each current cluster; wherein, the initial current cluster is the initial cluster, and the initial current cluster center is the initial cluster center. The current cluster corresponding to the largest weighted similarity measure value is selected as the target cluster. The current spectral data points are assigned to the target cluster to obtain the updated target cluster; The current cluster is updated based on the updated target cluster to obtain the updated cluster; Based on the updated cluster clusters, recalculate the corresponding updated cluster centers; Determine whether all spectral data points have been traversed; If the number of spectral data points contained in the updated cluster is greater than the preset number of points, the updated cluster will be used as the cluster centers. Otherwise, the updated cluster will be used as the current cluster for the next clustering operation, and the updated cluster center will be used as the current cluster center for the next clustering operation.

6. The transformer secondary winding fault detection method based on multispectral signals according to claim 4, characterized in that, The inter-turn corona discharge model, the surface discharge ablation model, and the suspended spark discharge model are set up in the following ways: Obtain defect-free transformer secondary windings; The insulation layer between two adjacent turns of enameled wire in the secondary winding of the defect-free transformer is scraped off to obtain an inter-turn corona discharge model. The insulation layer of continuous multi-turn enameled wire on the surface of the defect-free transformer secondary winding is scraped off to obtain a surface discharge ablation model. One turn of the coil in the defect-free secondary winding of the transformer is cut to obtain a floating spark discharge model.

7. A transformer secondary winding fault detection device based on multispectral signals, characterized in that, include: The system includes a multispectral data acquisition module, a discharge light pulse parameter calculation module, a light pulse amplitude judgment module, a fault-free judgment module, a fault judgment module, and a fault type identification module. The multispectral data acquisition module is used to acquire the first multispectral signal of the transformer to be tested; The first multispectral signal includes: a first spectral signal in several bands; The discharge pulse parameter calculation module is used to calculate the average discharge pulse amplitude and the proportion of discharge pulses in each band based on the first spectral signal of each band. The optical pulse amplitude judgment module is used to determine whether the average discharge optical pulse amplitude of all bands is less than the preset discharge amplitude. The fault-free determination module is used to determine, if yes, that the transformer under test is fault-free. The fault determination module is used to determine, otherwise, that the transformer under test has a fault. The fault type identification module is used to calculate the proportion of discharge pulses and the distance between preset cluster centers when it is determined that there is a fault in the transformer under test, and to determine the fault condition of the transformer under test based on the cluster center corresponding to the minimum distance. The fault conditions include: inter-turn corona discharge fault, surface discharge fault, or floating spark discharge fault.

8. The transformer secondary winding fault detection device based on multispectral signals according to claim 7, characterized in that, The preset cluster centers are determined in the following way: Acquire several sets of second multispectral signals of a preset fault model; wherein, the fault model includes: an inter-turn corona discharge model, a surface discharge ablation model, and a suspended spark discharge model; the second multispectral signal includes: second spectral signals of several bands; For each set of second multispectral signals for each fault model, the pulse count ratio of each band is determined based on the second spectral signal of each band, and the pulse count ratio of all bands in the second multispectral signal is used as the spectral data points of the second multispectral signal. Clustering is performed on all spectral data points of all fault models to obtain several cluster centers.

9. A terminal device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the transformer secondary winding fault detection method based on multispectral signals as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the transformer secondary winding fault detection method based on multispectral signals as described in any one of claims 1-6.

Images